Abstract: The dam temperature field has not been stabilized, and the time dependent effect does not increase non-monotonously for super high arch dams during initial operation period. Therefore a special deformation monitoring and forecasting model was developed, and its modeling method was proposed in the study. The key temperature measurement points were chosen by the hierarchical clustering on principal component method, and the corresponding time series were inputted as thermal predictors. The combined time dependent effect, including creep and its restoration, was introduced as time dependent deformation. This time dependent effect was employed to validate its characterization for the valley contraction deformation. Considering the predictor variables such as reservoir water level, adopted measured temperatures, estimated time dependent effect, the simple boosted regression tree (BRT) based dam deformation prediction model was constructed. Through the backward elimination method, a simplified BRT (SBRT) model only including major predictors was obtained. The marginal effects of variables on deformation were analyzed, and the relative influences can be quantitatively analyzed. With the help of partial dependence plot, the correlations among variables and the influences of variables on deformation can be explored, and the deformation mechanism can be revealed. The model was applied to a super high arch dam, and the case study verified the feasibility of the model. The results were compared with those by the support vector machine model and the traditional multiple regression models, which shows the superiority of the developed model.